Chromatic Pattern Recognition

摘要

An achromatic pattern recognition model is modified to process chromatic objects. The theory of color television is used to obtain pattern information in terms of one luminosity and two opponent chromaticity functions. Two-dimensional Fourier transformations of the two color functions are performed. Color identification is then made by correlation with transform prototypes of colored chips previously derived and stored. The object is identified in a similar manner by transforming its luminosity function. A digital simulation of the model is conducted using the English alphabet and 84 colors representative of the color spectrum. All letters and their colors are successfully identified, including conditions where pattern brightness varies. Characters were also identified when the chromaticity varied. It is concluded that the modified model can be trained to identify any object's color. (Author)